Ghostwriter MCP Server
Overview
Ghostwriter MCP Server is an executable Java 17 Model Context Protocol (MCP) server. It combines Ghostwriter's project-aware code-assistance workflows with Bindex Core's library-metadata retrieval, registration, and recommendation capabilities, and publishes those capabilities through the Machai MCP Server runtime. MCP clients, IDE integrations, and developer-automation processes can launch the assembled JAR locally over STDIO or connect to the server's HTTP mode.
The org.machanism.machai.mcp.server.McpServer entry point starts the unified server. With no port option it uses STDIO; --port enables HTTP mode, and --session enables streamable HTTP sessions. The server can also receive a project directory, server name, advertised version, and properties configuration through the runtime's command-line options. Its tools support project-aware assistance, model-assisted workflows, dependency metadata discovery, Bindex descriptor retrieval and registration, and library recommendations. The POM declares the Ghostwriter, Bindex Core, and Machai MCP Server runtime modules that provide these capabilities.
The source tree contains no package-info.java files. Accordingly, this overview is based on the Maven project description, declared dependencies, the executable entry point, and the project's Bindex metadata. Ghostwriter model-assisted workflows use the configured generative-AI provider; Bindex workflows use the configured metadata sources and registry services.
The diagram is maintained in src/site/puml/c4-diagram.puml. It shows the MCP client, Machai MCP Server runtime, Ghostwriter tools, Bindex Core services, the configured AI provider, and the external metadata and registry systems with which they interact.
Supported AI providers
AI-provider integration is supplied by the Machai Generative AI client used by Ghostwriter. CodeMie is the provider documented by this project and selected by its POM. Choose the provider with genai.serverId, use a provider-qualified model in gw.model, and supply credentials through runtime properties or environment variables rather than committing secrets to the POM. Other providers are not declared or documented in this module, so they are not listed as supported here.
| Provider | Status | Configuration |
|---|---|---|
| CodeMie | Project default; model-assisted workflows | Set genai.serverId=CodeMie and use a CodeMie model such as the default CodeMie:gpt-5.5-2026-04-24. Set GENAI_USERNAME and GENAI_PASSWORD; these may be a user name/password or client ID/client secret. Override AUTH_URL when using a non-default authentication endpoint. |
Common configuration parameters
| Parameter | Description | Default value in this project |
|---|---|---|
genai.serverId |
Generative-AI provider identifier used by Ghostwriter. | CodeMie |
gw.model |
Provider-qualified model identifier used for Ghostwriter model requests. | CodeMie:gpt-5.5-2026-04-24 |
GENAI_USERNAME / GENAI_PASSWORD |
CodeMie user credentials or OAuth client credentials. | No default; required for CodeMie authentication. |
AUTH_URL |
CodeMie OAuth token endpoint. | https://auth.codemie.lab.epam.com/realms/codemie-prod/protocol/openid-connect/token |
maven.compiler.release |
Java release used to compile the server artifact. | 17 |
--projectDir / -d |
Project directory supplied to project-aware server workflows. | Runtime-defined; commonly the current project directory. |
--port / -p |
Port that selects HTTP server mode. | Not set; no port starts STDIO mode. |
--session / -s |
Enables streamable sessions when HTTP mode is active. | Disabled. |
Maven properties can be overridden with -D options or in the Maven configuration used to build or run the server. Keep provider credentials out of the POM and source control. See the Machai MCP Server runtime documentation for the complete provider and command-line configuration reference.